Types and limits¶

Which column types COPY accepts, what it refuses, and how the extension is versioned.

Types COPY accepts¶

Per column of the spec:

Column type	Accepts
Text	`VARCHAR`, `ENUM` or `BLOB`
Integer	Any integer type, or a `FLOAT` or `DOUBLE` holding a whole number
Decimal	Any `DECIMAL` or integer type that holds the value exactly. Never `FLOAT` or `DOUBLE`.
Date	`DATE`, or `TIMESTAMP` at midnight
Time	`TIME` or `TIME_NS` in whole seconds

Tables and columns gives the type of every column.

A bare NULL inside a struct

DuckDB types a bare NULL inside a struct as INTEGER, and a text column refuses it. Write NULL::VARCHAR, or the column's type.

PARTITION_BY and PER_THREAD_OUTPUT ask for several files. DuckDB consumes them, and the format refuses a second file.
compression is refused too. Compress the file afterwards.

With DuckDB's default temporary file and a local path, an existing file is left untouched when the COPY fails.
Without it (USE_TMP_FILE false, or a path that is not local), the file is written over from its first byte.
In that case, an existing file longer than the new content is refused ("remove it first").
What happens to the target of a failed COPY is then up to DuckDB. Some versions remove it.

The extension is versioned separately from the Python package. Its version is in the crate's Cargo.toml.
Its releases are tagged duckdb-v*.
Its own changelog is crates/oxedi_duckdb/CHANGELOG.md.